On Thu, Mar 08, 2001 at 12:05:38PM -0800, Greywolf wrote:

 > Speaking of UVM/UBC, I notice that they have somewhat, if not completely,
 > edged out the page_zero_idle() loop...or has it?
 > 
 > I realise it's not trivial to keep a buffer cache and a cache of empty
 > pages available at the same time, especially on a busy system, but it
 > seemed to me that with that loop, having zeroed pages ready for processes
 > on demand, things were a bit snappier.

No, the reason that the idle-loop zeroing was disabled is because it was
adversely affecting performance with UBC.  The issue appears to be what
it does to the memory bus.

-- 
        -- Jason R. Thorpe <thorpej@zembu.com>